”More
Dynamic Range Day - Loudness War Protest

Production Advice

make your music sound great

Why mp3s suck, and how to hear it

by Ian Shepherd



I hate mp3, and this post will tell you why.

DO NOT read this post if you have a large collection of mp3s, enjoy listening to them and can’t hear any problems with them, because it’ll ruin them for you !

There’s been plenty written on how mp3 works, and why lossy compression sounds worse than uncompressed audio in general. My aim here is to demonstrate how mp3 sounds bad, for all the people who keep telling me there’s no difference.

I’m going to give you clear guidelines and examples on what to listen for and the negative effects of mp3, but there’s no going back – once you can hear the problems, you’ll never stop hearing them.

This isn’t limited to audiophiles, or “golden ears”, by the way – in my opinion anyone can hear this stuff, with a few pointers.

So seriously, unless you’re prepared to start using Ogg Vorbis, FLAC or AAC - stop reading now !

Still here ? Good.

First, I need to make this clear – I have nothing against lossy audio or data compression in itself – I do most of my casual listening on an iPod, using 128kbps AAC files – they sound fine. Not as good as the original CDs, obviously, but OK. And yes, I’m well aware that AAC is just a more advanced version of mp3. But the fact is that mp3 has fundamental limitations – even at higher bitrates.

Next – I’m also a pragmatist. mp3 is a temporary phenomenon, just like AM radio, cassettes and CDs. In the long run, none of those have killed music, and neither will mp3, or lossy compression in general. So, why the rant ?

Because people keep saying mp3 sounds great, or “indistinguishable from CD” and it’s just not true.

mp3 isn’t good enough

It doesn’t matter what encoder you use, it doesn’t matter what settings you use or what pre-processing you apply – mp3 just doesn’t cut it. AAC and later, more sophisticated encoders use more advanced encoding methods, and sound better to varying degrees, but mp3 just FAILs.

How does it fail ? That depends a little on the encoder being used, but some of my own pet hates include:

  • mp3 sizzle – the artificial, unnatural swirling metallic noises that sound like someone’s added chime bars to everything, or there’s a mosquito buzzing in your ear. Some people actually say we prefer these noises in mp3s – but the research says bullshizzle !
  • Added distortion – Yet another side effect of the so-called Loudness Wars. mp3 encoders rarely include any headroom for the encoding process itself, so the added processing pushes the music even further over the limits, generating inter-sample peaks and adding even more distortion in the process
  • Flat, two-dimensional sound mp3 works by throwing away musical information that we supposedly can’t hear – up to 90% of the original information, at 128kbps. That means all the subtle, delicate stuff, like ambience, space and realism. So a lush, three-dimension original is reduced to a flat, cardboard replica of itself
  • Mushiness All but the very best mp3 encodes just sound fuzzy, muddled and – well, mushy !


Hear for yourself

Don’t take my word for it – here are some examples. First, a truly nasty 128kbps mp3 example, from a Deep Purple live album I mixed a while back:

Please install Flash plugin

(Before anyone jumps on me, I’ve heard even a 256 kbps mp3s sounding like this – I’ve just used a low quality version to make the point.)

If that doesn’t sound too bad to you at first, try this - I’ve filtered the file to highlight the high frequencies. You can hear the problems most clearly when the vocals start:

Please install Flash plugin

Some people describe this effect as “sizzle”, or “swirlies”. It’s not just that I’ve removed all the bass, what I’m pointing out is the unatural bubbling, twinkling “chime-bar” type sound, or as my friend and fellow mastering engineer Nick Watson once called it, the “flocks of tweeting ultrasonic birdies”. It also reminds me of someone crinkling up tin foil !

Once you’ve picked it out, listen the first version again. Doesn’t sound so nice now, does it ? Can you ignore the swirlies, now you know they are there ?

Now download and listen to the original file:

‘Talk About Love – Excerpt’ – 5 MB WAV file

Listen to the clarity, punch, and bite of the WAV, compared to the swirly, soggy mess of an mp3. Which one do you prefer ?

The loss of depth, richness and three-dimensionality is more subtle side-effect, but just as unfortunate. Here’s a snippet of a recording I did for the brilliant Hans Koller, featuring Christine Tobin on vocals:

Please install Flash plugin

(This is a much better mp3 encode, with far fewer heinous swirlies. But still…)

Here’s the WAV version:

‘The Great Bear And The Small – Excerpt’ – 11 MB WAV File

Don’t expect the difference here to leap out at you straight away, it’s more a case of feeling it – listen to the swirls of the harp from 30 seconds in, listen to the piano and Christine’s voice – on the wav file, there’s a warmth, and a depth, and a sparkle that in the mp3 has just gone.

Listen to the wav several times over, then switch to the mp3. Do you honestly feel it sounds as good ? The mp3 is OK, but it’s just… meh. I’m not drawn in, my attention wanders, it doesn’t move me.

Something essential has been lost, and you can’t get it back. And once you’ve heard that loss, even cranking the data-rate up doesn’t help. The only solution is a more advanced format, or lossless files.

Try listening to the mp3s in your music collection. Go back and compare them to the CDs you ripped them from.

…Sorry.

Update:

I’ve had lots of interest in this post, and lots of discussion, especially on link-sharing sites. There are a few common responses that I want to answer here.

No-one uses 128 kbps mp3s

Wrong. If you’ve made this comment, you probably already know about LAME and the all other flavours of mp3 codec, and you probably do choose to use higher bit-rates, but you’re in the minority. Most “regular listeners” go for the default settings – and even in iTunes this is only 160 kbps.

192/320 kbps sounds fine

Sometimes. This depends so heavily on the material, the encoder and the codec – you simply can’t make blanket assumptions. Ironically one of the factors that makes mp3 so popular – the fact that there are so many encoders and players, some of which are free – also makes it far harder to get a decent encode. By contrast, the grip Apple have over the AAC format at least ensures consistently high standards of encoding.

You’re just an Apple fanboy

No. Well alright, yes – I am a big fan of Apple’s products, but there are plenty of other alternatives to mp3 – OGG Vorbis, FLAC etc. The only reason I mention AAC a lot is it’s a format I have deep experience of, and always sounded good (but not perfect !) to me.

And another thing

To everyone who keeps saying “just use 320 kbps”, I say – why ?!? mp3 simply has inherent limitations compared to other formats. The whole point of lossy audio is to save space. At 128 kbps that saving is 90% – well worth having. At 320 kbps though, that saving is only 60% and it still doesn’t sound great – I’d far rather go with FLAC or Apple lossless, which can often achieve an almost equivalent 50% saving in space, and have something that sounds identical to the source.

Update #2 (2014):

Five years later, and the mp3 (and lossy data compression in general) is still with us.

AAC sounds even better since Apple’s Mastered for iTunes initiative, and even offers tantalising glimpses of a lossless future. Meanwhile Neil Young has his heart in the right place with his Pono initiative – but I have some concerns. And Harman seem to regard the whole thing as nothing more than a marketing opportunity.

For another chance to hear the problem with 128 kbps mp3s for yourself, try this test:

http://mp3ornot.com

It’s only 128 kbps versus 320 kbps unfortunately, but it’s an interesting exercise all the same.

Here’s how I did:

mp3 test results

Let’s just go lossless and be done with it !


Image by Roger B

facebook comments:

203 Responses

  1. steve says:

    The MP3 samples provided are very poor quality even for 128kbps. Feel free to check the audio with Sonic Visualiser or a similar spectrum analyzer and you’ll see for yourself. The sample MP3s don’t look anything like the original and if you encode them yourself, they’ll at least look like the original file. It appears that someone has messed with the sample to make it sound worse than the original and you could most likely tell the difference even before compressing it.

    Samples on the mp3ornot site (it’s still mp3 regardless of bitrate) are also in horrible quality, at 320kbps they’re even worse than those provided here at 128kbps. Not to mention the insane amount of dynamic range compression in the samples.

    I’m personally using high bitrate AAC and FLAC, but if you think that MP3 sounds this bad… I have some sad news for you friend, it’s not because you have magic ears of the audio Gods :)

  2. Ian Shepherd says:

    Hi Steve,

    I made the samples myself and I absolutely guarantee that they haven’t been “messed with”. Full disclosure – I wanted an example most people could hear so the Deep Purple example was chosen deliberately because it gave a poor result. However second clip is a very fair representation, in my experience.

    I know I don’t have magic ears, and I’m completely comfortable with that – in fact that’s the problem, the shortcomings of mp3 are audible to anyone !

    Feel free to post examples you think show mp3 in it’s best possible light, but I’m entirely confident I still won’t like it.

    Ian

  3. J7N says:

    These samples are easy to listen to. They are quite poor quality; seems that ’3TalkAboutLoveAmLow’ was encoded with -q 9 which the fastest setting. The rock concert has the drums smeared of course. And the voice has acquired some properties of the whistling guitar, and in the vocal sample the singer occasionally sings the harp, as if those signals were “modulating” the voice.

    I am not good at hearing Mp3 in music. Where compression is easy to pinpoint is human voice and solo acoustic instruments. Breath sounds, and ‘sss’ that should be disorderly, unpredictable, smooth, become spiky, rolling, watery. The same effect can be heard on denoised old recordings, which I am bothered by, due to collapse and dropout of frequency bands.

    I normally listen to FLAC to get “a piece of mind” that the occasional ringing spikes I hear are not a fault, and were intentionally mixed in the album. I have a few albums released starting from mid-nineties that sound electronic and phasey, and are easy suspects for MP3. Basically the vocals sound like Goa’uld or the Combine. I fail to precisely describe the sound as I only have a vague understanding of audio mixing. I’ve cut a few samples of those records. Would you listen and express your opinion?

    Thank you for the link to the Brandenburg paper. Point 5.2 talks about dynamic range. Maybe it is that encoders not the format can still be tested and described. LAME will not encode extremely quiet sounds in VBR mode (for example, film soundtrack), and the CLI encoder will clip values above full scale. (CBR, and the lame_enc.dll, thus have the widest dynamic range). The paper also mentions voice as a good test material. I guess because of dynamics and low possibility of masking compared to music that is often loud across the spectrum.

    Regarding the of low bitrates, I’ve an anecdotal evidence. I used to know a sound engineer who was expert in old tape recorders and amplifiers and other solid old school tehcnology. When we met he gave me a lecture how the MiniDiscs I was to use operated in floating point precision (obviously it’s quantized), and thus were better. He also used 128 kBit/s to encode music and compilations for personal use. He cited an audio bandwidth that this bitrate was supposed to support (it’s actually choosable as per 5.7.2), and would not listen to me when I suggested the dolby standard bitrate of 192k as a starting point. As far as I am aware, he never had a modem to use “web quality” with.

    I am looking forward to wider adoption of OPUS, as it addresses the band collapse issue that irritates me, and conceals it with noise synthesis. It’s a compact format with simple, flexible tagging, directly compatible with FLAC, Vorbis and APEv2 (multitude of formats), thus avoiding the issue of losing tags when translating to format specific frames in ID3 and MP4. One curious issue I observed that Opus, unlike other modern formats, does clip at +6 dB. Maybe the useful range was carefully limited for efficiency.

Leave a Reply

Ian Shepherd


BBC Radio 4 Interview

Please install Flash plugin

Ian Shepherd from Production Advice discusses the Loudness Wars

Connect